Bearing raceway grinding process parameter optimization method based on hybrid machine learning model

Abstract

The invention discloses a bearing raceway grinding process parameter optimization method based on a hybrid machine learning model, and the method comprises the steps: constructing a data set of bearing raceway grinding process parameters, and dividing the data set into a training set A and a test set; a hybrid machine learning model is constructed and trained, and the hybrid machine learning model is used for outputting predicted performance index parameters after the grinding process parameters are input; and optimal process combination parameter combination solving is achieved based on a TSO-NSGA-II algorithm, specifically, for grinding of a machining target, the TSO-NSGA-II algorithm is adopted, a predicted value of the hybrid machine learning model serves as a fitness function to conduct parameter optimization till process parameters with the optimal result of the machining target are obtained, and the machining target is ground according to the process parameters. The hybrid machine learning model provided by the invention not only can be used in the aspect of a bearing raceway grinding scene, but also can be used in other similar machining process optimization scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of raceway grinding, and in particular to a method for optimizing process parameters of bearing raceway grinding based on a hybrid machine learning model. Background Art

[0002] Bearings are common key components in mechanical devices. They are composed of inner rings, outer rings, rolling elements and cages. They are used to reduce friction in rotating machinery and support the weight of bearing parts. High-performance bearings play a key role in major equipment and high-precision equipment. However, my country's independent research and development capabilities in high-speed, heavy-load and high-precision bearings still lag behind the international advanced level. The high-end bearing market has long been dominated by foreign countries, which has seriously restricted the development of my country's equipment manufacturing industry. Therefore, improving the design and manufacturing level of rolling bearings and improving the manufacturing process have become issues that need to be urgently addressed in my country's bearing industry.

[0003] As an important surface in the working process of the bearing, the surface quality and dimensional accuracy of the bearing raceway will directly affect the service performance and fatigue life of the bearing. Therefore, the raceway grinding process is crucial to the quality of the bearing. Improper selection of grinding parameters will lead to a large fluctuation range of grinding force during the processing process, which will have an adverse effect on the surface quality and processing accuracy of the ring, which will continue to the finished bearing, and ultimately reduce the service life and service performance of the bearing. In the actual processing and production process, there are corresponding requirements and standards for the surface quality and dimensional accuracy of the workpiece. In order to ensure qualified or even excellent workpiece surface quality and dimensional accuracy, the grinding process parameters are generally adjusted continuously at the production site, and the surface quality and dimensional accuracy of the workpiece after processing are continuously measured, which generates a lot of repetitive work, resulting in low efficiency, and this method has a certain degree of hysteresis. In order to complete the production and processing work with higher production and processing quality and efficiency, it has become a problem to be solved urgently to find the regular relationship between the workpiece grinding process parameters and its surface quality and dimensional accuracy, and to achieve rapid and accurate prediction of them.

[0004] Patent: CN115600504A A method for optimizing grinding parameters of CNC grinders based on an improved reinforcement learning algorithm first obtains the grinding specific energy formula and the material surface roughness formula, then uses the multivariate linear regression and least squares method principles to fit the coefficients that affect the grinding specific energy and material roughness, and establishes a mathematical model for optimizing the processing parameters of the grinding stage of the CNC grinder; the grinding parameters are optimized by using multiple intelligent agents to learn from each other, and then the optimal grinding parameters are obtained.

[0005] However, further research shows that the above technology has the following deficiencies and defects:

[0006] 1. The actual grinding process may be affected by a variety of factors that are difficult to quantify. The mathematical model may not be able to fully cover all actual conditions, resulting in a certain deviation between the optimization results and the actual situation.

[0007] 2. The uncertainty of multi-agent learning. The way of mutual learning may have instability or convergence problems, and it may not be possible to reliably obtain the optimal parameters in complex scenarios. Summary of the invention

[0008] Technical purpose: In view of the defects in the prior art, the present invention discloses a method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model.

[0009] Technical solution: In order to achieve the above technical objectives, the present invention adopts the following technical solution.

[0010] A method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model, the method comprising:

[0011] Construct a data set of bearing raceway grinding process parameters and divide it into training set A and test set;

[0012] Construct and train a hybrid machine learning model, which is used to input grinding process parameters and output predicted performance indicator parameters;

[0013] The optimal process combination parameter combination is solved based on the TSO-NSGA-II algorithm: For the grinding of the processing target, the TSO-NSGA-II algorithm is used, and the predicted value of the hybrid machine learning model is used as the fitness function to optimize the parameters until the optimal process parameters of the processing target are obtained, and the processing target is ground according to the process parameters.

[0014] Beneficial effects:

[0015] (1) The present invention proposes a hybrid learning-based model, which improves the predictive performance of the basic model by establishing SVR, RF, GBDT, ADaBoost, and LR basic models, and by using grid search optimization and cross-validation methods. On this basis, based on the Stacking idea, feature engineering is performed, and the SVR, RF, GBDT, and ADaBoost models are used as primary learners, and the LR model is used as a secondary learner, which can significantly improve the overall prediction performance, which is better than using a single model alone, and reduces the risk of overfitting, enhances a certain nonlinear generalization ability, and improves the prediction accuracy of the target processing results;

[0016] (2) The present invention can conveniently and quickly combine as many relevant experimental results as possible, the amount of data is more sufficient, the influence of subjective factors is reduced, and the credibility of the prediction results is improved;

[0017] (3) The TSO-NSGA-II algorithm proposed in the present invention obtains a series of Pareto optimal solutions, and then determines the final grinding process parameters from a series of Pareto optimal solutions according to actual requirements through the principal component analysis method, thereby obtaining the optimal surface roughness and roundness of the bearing raceway.

[0018] (4) The hybrid machine learning model proposed in the present invention can be used not only in the bearing raceway grinding scenario, but also in other similar processing technology optimization scenarios. On the basis of doing minimal experiments, the optimal processing parameters in other scenarios can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention;

[0020] Figure 2 This is the architecture diagram of the Stacking integrated learning model of the present invention;

[0021] Figure 3 The Stacking training flow chart of the present invention;

[0022] Figure 4 This is a flow chart of the TSO-NSGA-II algorithm of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] As attached Figure 1 As shown, a bearing raceway grinding process parameter optimization method based on a hybrid machine learning model of the present invention comprises the following steps:

[0025] S1. Construct a data set of bearing raceway grinding process parameters and divide it into a training set A and a test set; including the following steps:

[0026] S11, collecting data corresponding to several groups of grinding process parameters and performance index parameters; selecting several bearing raceway grinding process parameters and performance index parameters, and performing several bearing raceway grinding experiments on the bearing workpiece based on the grinding process parameters to obtain corresponding performance index parameters; obtaining the parameter range of each grinding process parameter, and performing several bearing raceway grinding experiments on the bearing workpiece based on the orthogonal combination arrangement of the grinding process parameters to obtain corresponding performance index parameters;

[0027] In some embodiments of the present invention, the grinding process parameters include the workpiece axis speed n g , Grinding wheel spindle speed n s , Grinding wheel dressing speed n x , coarse grinding speed v c and fine grinding speed v j ; Performance index parameters include surface roughness R a and roundness Cir; obtain the parameter range of each grinding process parameter according to experience or other methods, select within the parameter range of the grinding process parameter, and conduct multiple groups of bearing raceway grinding experiments through the orthogonal arrangement method to obtain the surface roughness R a and roundness Cir performance index parameter data; in each experiment, multiple different positions on the bearing workpiece can be evenly selected, such as 5 different positions for surface roughness R a The average value of the roundness Cir is taken as the surface roughness R under the current grinding process parameter combination. a And the value of roundness Cir.

[0028] The level value of the grinding process parameter is preferably x. The level values ​​of all grinding process parameters are combined according to the orthogonal experiment method. Five level values ​​are selected for each grinding process parameter to facilitate subsequent combination experiments. In addition, each raceway grinding experiment is performed on the surface roughness R a The values ​​of Cir and roundness are measured multiple times, and the average value is taken as the experimental result under different level value combinations.

[0029] S12. Clean the collected data to obtain cleaned data. In some embodiments of the present invention, for the values ​​of the acquired performance indicator parameters, an average value is taken based on the performance indicators of the five different positions taken above, and then a relative error analysis is performed between the five values ​​and the average value, and data exceeding 10% is cleaned.

[0030] S13, performing dimension normalization processing on the cleaned data to obtain a data set of bearing raceway grinding process parameters, and dividing the data set into a training set and a test set. The division ratio of the training set A and the test set is preferably set to m:n; in some embodiments of the present invention, the division ratio is set to 8:2.

[0031] In order to avoid the impact of the dimensional differences between different types of parameters on the subsequent machine learning model training, each type of parameter can be dimensionally normalized, that is, each type of grinding process parameter and performance index parameter is dimensionally normalized. The calculation formula for dimensional normalization is:

[0032]

[0033] Among them, x is the original data, that is, the data after cleaning, x min is the minimum value in the original data, x max is the maximum value in the original data, is the dimension-normalized value.

[0034] S2. construct and train a hybrid machine learning model, where the hybrid machine learning model is used to input grinding process parameters and output predicted performance index parameters;

[0035] As attached Figure 2 and attached Figure 3 As shown, the hybrid machine learning model is to train T different basic models based on the stacking integration idea, select the first number of basic models as the first layer prediction model of stacking learning, select the LR basic model as the second layer prediction model of stacking learning, and the first layer prediction model and the second layer prediction model of stacking learning are used as the hybrid machine learning model; T is the sum of the number of first layer prediction models and second layer prediction models. The scheme shown in the embodiment of the present invention is the optimal scheme selected in the actual verification process, that is, the model algorithm based on the SVR, RF, GBDT, and AdaBoost basic models as the base learner layer, that is, the first layer prediction model, and then according to the new data set generated by the first layer prediction model, the LR basic model is selected in the second layer data model to obtain the final prediction result; the content includes:

[0036] Select multiple different basic models, including SVR, RF, GBDT, ADaBoost, BP neural network, ridge regression, XGBoost, and LightGBM as the first-layer prediction model of stacking learning, and the LR basic model as the second-layer prediction model of stacking learning, and use training set A to train them. First, use training set A to train the first-layer prediction model of stacking learning; then use the test set to evaluate the prediction performance of all models in the first-layer prediction model of stacking learning, select the model with higher prediction accuracy as the first-layer prediction model of stacking learning, and combine the LR basic model as the second-layer prediction model of stacking learning to obtain a hybrid machine learning model; then divide training set A into training set B and validation set, and finally find the optimal hyperparameters for the hybrid machine learning model to improve its prediction accuracy. Then use the selected basic model as the base learner and meta-learner of the stacking ensemble idea to establish the surface roughness R a and the prediction model of circularity Cir. The split ratio of the training set B and the validation set is preferably set to m:n;

[0037] The present invention trains basic models through data sets, selects feature engineering, and establishes five basic models, namely SVR, RF, GBDT, ADaBoost, and LR. During each training, grid search and 5-fold cross validation are performed on the training data to find the optimal hyperparameters, and the RF algorithm is used to evaluate the importance of the features, so that each basic model uses the optimal hyperparameters. Then a stacking ensemble model is established, and RMSE and MAE are used to evaluate the basic models and stacking models.

[0038] The training process of the hybrid machine learning model includes:

[0039] The 5-fold cross-validation method is introduced to train each base learner, and the training set A is evenly divided into 5 mutually exclusive subsets, namely A1, A2, A3, A4, and A5. Then the union of 4 of the subsets is selected as the training set, and the remaining 1 subset is selected as the test set to construct the training set and test set of the primary learner. In this way, each primary learner can obtain 5 sets of training sets and test sets, and obtain 5 test results S1, S2, S3, S4, and S5.

[0040] Because the Stacking ensemble prediction contains 4 base learners, these 5 training sets and test sets will be used to train and test the base learners. For the nth base learner, after completing the 5-fold cross-validation, there will be 5 prediction results, which are stacked vertically by row to obtain the prediction set S of the sample data under the base learner. i，n (i=1, 2, 3, 4, 5), and the average of these 5 prediction results is After completing the training of the four base learners, the prediction sets and prediction means obtained by each base learner are concatenated column by column to obtain the new training set S of the second layer. i，n , with the new test set Form the input training set of the second layer learner. That is, the data set

[0041] Use the original learner LR for secondary training. Input the new training set and test set obtained by the previous step of primary learner training into the second-layer meta-learner for secondary training to obtain the final surface roughness R a And the prediction results of circularity Cir.

[0042] The established prediction model needs to use evaluation indicators to evaluate the prediction results of SVR, RF, GBDT, AdaBoost, LR model and stacking integrated model, that is, to determine its prediction accuracy. The calculation formulas of the relevant evaluation indicators RMSE and MAE are as follows:

[0043]

[0044] in, represents the true value in the test set, Represents the predicted value output by the prediction algorithm; m is the number of experiments, that is, the number of data in the data set constructed in S1.

[0045] S3, based on TSO-NSGA-II algorithm to achieve the optimal process combination parameter combination solution;

[0046] For the grinding of the processing target, the TSO-NSGA-II algorithm is used, and the predicted value of the above hybrid machine learning model is used as the fitness function to perform parameter optimization until the optimal process parameters of the processing target are obtained, and the processing target is ground according to the process parameters.

[0047] The optimized process parameters described in the TSO-NSGA-II algorithm are optimized by selecting relatively important process parameters based on feature importance evaluation. In addition, the results obtained need to be optimized from a series of Pareto solutions through principal component analysis.

[0048] The specific steps are as follows:

[0049] S31. First, design the decision variables and select the workpiece axis speed n according to the feature importance evaluation results of the random forest algorithm, that is, the RF algorithm. g , Grinding wheel spindle speed n s , Grinding wheel dressing speed n x , coarse grinding speed v c and fine grinding speed v j The three most important process parameters among these five characteristic parameters are used as decision variables of the multi-objective optimization model.

[0050] S32, then impose constraints, according to the process parameter selection range recommended by the operator, impose constraints on the decision variables, and then establish the surface roughness R a The objective function of the multi-objective optimization model is the minimum surface roughness and the minimum roundness. The objective function calculation formula is:

[0051]

[0052] in, The surface roughness R predicted by the proposed hybrid machine learning model a and the value of roundness Cir, ω1 is The weight coefficient, ω2 is The weight coefficient is ω1+ω2=1; the value is set according to the actual situation. If the two are of equal importance, both are taken as 0.5.

[0053] S33. Based on TSO-NSGA-II, the grinding process parameters are optimized to obtain several optimal Pareto solution sets. The parameter settings of the TSO-NSGA-II algorithm need to be continuously adjusted according to the optimization effect until the best optimization effect is found. Figure 4 As shown in the figure, the grinding process parameter optimization based on TSO-NSGA-II includes the following steps:

[0054] S33 1. Use the tuna algorithm to initialize the population Q, and the number of iterations is set to 1; the individual value is the possible optimal solution of the five process parameters, that is, the characteristics (decision variables) of the tuna group individuals are used as the individuals of NSGA2.

[0055] S332. Evaluate the objective function value of each individual in the tuna algorithm;

[0056] S333, calculating the congestion degree by fast non-dominated sorting method in tuna algorithm;

[0057] S334, use NSGA2 algorithm to perform selection, crossover and mutation, generate offspring, and increase the number of iterations by 1;

[0058] S335, update offspring using tuna algorithm;

[0059] S336, evaluating the objective function value of the offspring in the tuna algorithm;

[0060] S337, merge parent and offspring populations;

[0061] S338. Calculate the congestion degree by non-dominated sorting method in tuna algorithm;

[0062] S339, generate a new generation of population;

[0063] S340, determine whether the termination condition is met, if yes, then end, if not, return to S334 to regenerate children; the termination condition is the maximum number of iterations, the maximum number of iterations is user-defined, in some embodiments of the present invention, the maximum number of iterations is set to 1000. Finally, a series of optimal Pareto solution sets are obtained.

[0064] S34. Determine the optimal process parameters based on principal component analysis for the optimal Pareto solution set. The principal component analysis method is used to further determine the final optimal process parameters; the calculation formula of the principal component analysis method is:

[0065]

[0066] The score of each group of non-dominated solutions in all optimal Pareto solutions is calculated according to the formula, where S i is the score of the i-th group of process parameters, Y is the principal component vector, w iThe variance contribution rate of the principal component. The calculation process of the variance contribution rate of the principal component is: after the principal component analysis, its eigenvalue can be obtained, and the eigenvalue is divided by the sum of all eigenvalues ​​to obtain the variance contribution rate of the principal component. The result can be directly generated by SPSS software.

[0067] After calculating the principal component scores, sort the total scores of all solutions from high to low. Then, according to the actual production requirements, the solution with the smallest surface roughness and the smallest roundness can be selected, and then it is the optimal solution for the optimization of bearing raceway grinding parameters.

[0068] The present invention also discloses an electronic device, which includes: a memory for storing a computer program; and a processor for executing the computer program so that the device executes the aforementioned method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model.

[0069] The present invention also provides a computer storage medium, on which a computer program is stored. When the computer program is executed, the device running the computer program implements the aforementioned method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model.

[0070] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, and the computer software product can be stored in a storage medium, and the memory can be various types of memory, which can be a random access memory, a read-only memory, a flash memory, etc., such as a read-only memory (English: read-only memory, ROM) / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0071] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model, characterized in that: The method comprises: Construct a data set of bearing raceway grinding process parameters and divide it into training set A and test set; Construct and train a hybrid machine learning model, which is used to input grinding process parameters and output predicted performance indicator parameters; The optimal process combination parameter combination is solved based on the TSO-NSGA-II algorithm: For the grinding of the processing target, the TSO-NSGA-II algorithm is used, and the predicted value of the hybrid machine learning model is used as the fitness function to optimize the parameters until the optimal process parameters of the processing target are obtained, and the processing target is ground according to the process parameters.

2. The method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model according to claim 1, characterized in that: Construct a data set of bearing raceway grinding process parameters, including: Collect data corresponding to several groups of grinding process parameters and performance index parameters; Perform data cleaning on the collected data to obtain cleaned data; The cleaned data were dimensionally normalized to obtain a data set of bearing raceway grinding process parameters.

3. The method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model according to claim 2, characterized in that: Collecting data corresponding to several groups of grinding process parameters and performance index parameters, including: selecting several bearing raceway grinding process parameters and performance index parameters, and performing several bearing raceway grinding experiments on the bearing workpiece based on the grinding process parameters to obtain corresponding performance index parameters; obtaining the parameter range of each grinding process parameter, and performing several bearing raceway grinding experiments on the bearing workpiece based on the orthogonal combination arrangement of the grinding process parameters to obtain corresponding performance index parameters.

4. The method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model according to claim 1, characterized in that: Hybrid machine learning models include: Based on the stacking ensemble idea, T different basic models are trained, the first number of basic models are selected as the first-layer prediction models of stacking learning, the LR basic model is selected as the second-layer prediction model of stacking learning, and the first-layer prediction model and the second-layer prediction model of stacking learning are used as hybrid machine learning models; T is the sum of the first-layer prediction models and the second-layer prediction models.

5. The method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model according to claim 1, characterized in that: The TSO-NSGA-II algorithm is used to optimize the parameters of the hybrid machine learning model using the predicted value as the fitness function, including: Design decision variables, select the workpiece axis speed n g , Grinding wheel spindle speed n s , Grinding wheel dressing speed n x , coarse grinding speed v c and fine grinding speed v j Three process parameters among these five characteristic parameters are used as decision variables in the multi-objective optimization model; Constraints are imposed on the decision variables, and then the surface roughness R is established. a And the objective function of circularity Cir; Based on TSO-NSGA-II, the grinding process parameters are optimized and several optimal Pareto solution sets are obtained; The optimal process parameters are determined based on principal component analysis of the optimal Pareto solution set.

6. The method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model according to claim 5, characterized in that: The objective function calculation formula is: in, and Surface roughness R predicted output for hybrid machine learning model a and the value of roundness Cir, ω1 is The weight coefficient of The weight coefficient of .

7. The method for optimizing bearing raceway grinding process parameters based on a hybrid machine learning model according to claim 5, characterized in that: The grinding process parameter optimization based on TSO-NSGA-II includes the following steps: S331, using the tuna algorithm to initialize the population Q, the number of iterations is set to 1; the individual value is the possible optimal solution of the five process parameters; S332. Evaluate the objective function value of each individual in the tuna algorithm; S333, calculating the congestion degree by fast non-dominated sorting method in tuna algorithm; S334, use NSGA2 algorithm to perform selection, crossover and mutation, generate offspring, and increase the number of iterations by 1; S335, update the offspring using the tuna algorithm; S336, evaluating the objective function value of the offspring in the tuna algorithm; S337, merge parent and offspring populations; S338. Calculate the congestion degree by non-dominated sorting method in tuna algorithm; S339, generate a new generation of population; S340, determine whether the termination condition is met, if so, end, if not, return to S334 to regenerate the offspring; the termination condition is the maximum number of iterations, and finally a series of optimal Pareto solution sets are obtained.

Images